The SaaS Dream, Interrupted
A few months ago, we were heads-down building PromptMetrics as a traditional SaaS. It was going to be the ultimate EU-focused, privacy-first LLM observatory. We had the architecture mapped, the pipelines built, and the pricing tiers set. We were ready to scale.
Then, the floor fell out from under us.
An employer missed a mandatory insurance renewal. That single administrative error triggered a cascade of consequences: work permit renewals were blocked, permanent residency plans were canceled, and our Swedish company had to be dissolved. Just like that, the foundation we'd built in Malmö evaporated.
We relocated to Berlin for a fresh start: same team, same vision, new network.
Then we hit the next wall: German regulations barred us from self-employment for a full year and from incorporating a GmbH until 2027. Most teams would have paused, packed it up, or waited it out. We almost did.
But while we were stuck in regulatory limbo, we started talking to the community.
The Pattern We Couldn't Ignore
Every CTO we met was grappling with the same problem. They were running LLMs in production, handling sensitive user data, and desperately needed observability they could actually trust.
They didn't want a black-box dashboard routing their prompts through US-based infrastructure. They didn't want another YC-backed tool priced for Silicon Valley burn rates. They needed a solution built for EU data sovereignty, GDPR-by-default, and the reality of running models on European servers.
The existing options on the market were always some combination of:
- Too expensive: Priced out of reach for bootstrapped teams or companies not burning VC cash.
- Too complex: Over-engineered for teams that need clear visibility, not a second full-time job managing the tool.
- Too Silicon Valley: Misaligned with European data philosophies in their architecture and pricing models.
We heard this from healthcare startups in Berlin, fintech teams in Amsterdam, and government contractors in Paris. The need was glaring, and nobody was serving it correctly.
The Realization
We already had the codebase. We already had the architecture. What we didn't have was a reason to rip the paywall off and let the community stress-test it.
The visa disaster gave us that reason.
Without a legal entity to sell through, we couldn't run a traditional SaaS even if we wanted to. But we could publish the code. We could let teams self-host. We could build in public and let the engineers who actually needed this tool shape its future.
So we did.
What PromptMetrics Actually Does
At its core, PromptMetrics is an LLM observability platform built strictly for teams that care about where their data lives.
- 100% Self-Hostable: Deploy on your own infrastructure, including EU data centers.
- Zero Prompt Exfiltration: Your data never leaves your environment for third-party analytics.
- Real-Time Tracing: Monitor latency, token usage, costs, and error rates across multiple providers in real time.
- Local Evaluation Pipelines: Run evals on your own hardware, not in someone else's GPU cluster.
- Open Standards: Fully OpenTelemetry-compatible traces, making it seamlessly exportable to your existing observability stack.
It's designed for the team running Ollama on a Hetzner server, calling Mistral through a French API endpoint, or just trying to monitor their OpenAI usage without shipping their entire payload to LangSmith.
Open Source by Design, Not Default
There's a tendency in tech to frame open-sourcing as the fallback plan when a business model fails. That isn't the case here.
Being forced into regulatory limbo gave us a rare do-over: the chance to build PromptMetrics the way it should have been built from day one. Transparent, community-driven, and perfectly aligned with EU infrastructure needs. A paywall was always going to limit who could use this; removing it didn't shrink our ambition, it expanded our surface area.
We aren't abandoning a business model, we're choosing a better one. One where the gold standard isn't what VCs will fund, but what engineers will actually trust in production.
What's Next
The repository is live today. We're launching with our core observability engine and evaluation framework, and our roadmap is completely public.
If you are running LLMs in production and want deep observability without surveillance capitalism, we want your feedback, bug reports, and pull requests.
We're also actively speaking with European hosting providers and infrastructure funds about sustainable funding models that don't rely on selling user data or chasing hypergrowth. If you're working on EU digital sovereignty and see alignment, our DMs are open.
Try It Out & Contribute
Whether you want to deploy it for your team or help us build the next big feature, you can get started in seconds:
git clone [https://github.com/iiizzzyyy/promptmetrics](https://github.com/iiizzzyyy/promptmetrics)
cd promptmetrics
docker compose up
Read the full docs and check out our open issues at: https://github.com/iiizzzyyy/promptmetrics
PromptMetrics is officially open-source, and we want you to be part of it.
We are actively looking for contributors! Whether you're a seasoned engineer wanting to add new model integrations or a first-time open-source contributor looking to help with documentation or squash some bugs, your pull requests are deeply welcome.
Help us build the standard for LLM observability in Europe.
— Berlin, April 2026
Top comments (0)